Disease named entity recognition from biomedical literature using a novel convolutional neural network

Area of Science:

• BioNLP
• Computational Linguistics
• Biomedical Informatics

Background:

• Automatic disease named entity recognition (DNER) is crucial for advanced BioNLP tools.
• Conventional DNER systems often require extensive, time-consuming feature engineering.
• Existing deep learning methods can be complex due to the inclusion of Conditional Random Field (CRF) layers.

Purpose of the Study:

• To propose a novel Multiple Label Convolutional Neural Network (MCNN) for disease NER.
• To introduce a Multiple Label Strategy (MLS) as an alternative to CRF layers.
• To reduce the need for manual feature engineering in DNER systems.

Main Methods:

• A MCNN architecture is developed for disease NER.
• Character, word, and lexicon embeddings are concatenated.
• Convolutional layers are applied, followed by the proposed MLS at the output layer.
• The MLS captures correlations between neighboring labels without a CRF layer.

Main Results:

• The MCNN approach achieves state-of-the-art performance on the NCBI and CDR datasets.
• Experimental results validate the effectiveness of the MLS strategy.
• The system demonstrates strong performance with minimal feature engineering.

Conclusions:

• The proposed MCNN-based disease NER method achieves state-of-the-art performance.
• The MLS strategy effectively captures label correlations in neighboring contexts.
• This approach offers a more efficient and effective solution for disease NER.